Empirical Investigation of the Predictive Validity of Machine Learning Models for Graduate Employability in Nigerian Universities

Graduate employability prediction using university-collected data can inform student support interventions and curriculum decisions, yet the predictive validity of machine learning models for this outcome in Nigerian higher education contexts has not been empirically evaluated. This study empirically tested the predictive validity of machine learning models for graduate employability within 12 months of graduation using alumni follow-up data from two Nigerian federal universities: University of Ibadan (South West) and University of Jos (North Central). A combined dataset of 4,200 graduates across three graduation cohorts was constructed from university registry records and an alumni tracer survey. Features included final CGPA, faculty, internship participation, NYSC posting type, extracurricular leadership, scholarships, and socioeconomic proxy indicators. Logistic regression, random forest, and neural network models were trained on an 80-20 stratified split. Random forest achieved the highest AUC of 0.81 and an F1-score of 0.77 for the employed class. Internship participation was the strongest individual predictor across all models. Faculty membership showed significant zone-level interaction, with Engineering graduates in Jos showing better outcomes than University of Ibadan counterparts, contrary to institutional rank assumptions. CGPA was a moderate but consistently significant predictor. The study fills a gap in the Nigerian graduate employability modelling literature and recommends universities use model outputs to target internship facilitation, and that NUC incorporate employability prediction infrastructure into its institutional performance monitoring framework.

Keywords: graduate employability prediction, machine learning, Nigerian universities, alumni data, predictive validity